What a Google Ads Manager Gets Wrong in Their First ChatGPT Campaign
By Keshav Parsai 8 min read
In brief
Avoid five Google Ads habits that weaken a first ChatGPT Ads campaign, from keyword-style structure to premature bid and budget changes.
Last verified: 12 September 2026 | Version: 1.0 | Next scheduled review: 12 October 2026
Nothing in the list below is incompetence. Each one is a correct Google Ads instinct applied where its supporting mechanism is missing. The failure mode is not ignorance of the new platform, it is fluency in the old one.
Five habits, five corrections.
Habit one: building granular ad groups
What the instinct says. Tighter ad groups mean tighter ad-to-query match, better quality score, lower CPC. In the extreme this becomes SKAG, one keyword per ad group, and it worked for years on Google.
Why it fails here. ChatGPT Ads structure is Campaign > Ad group > Ad. Reporting stops at the ad group. There is no unit below it that returns numbers: no keyword rows, no search terms, no per-hint breakdown. When you build fourteen ad groups on a 25 dollar daily minimum, you have not increased precision, you have split one readable dataset into fourteen unreadable ones.
There is also no quality score to optimise toward. Relevance is weighted in the auction but never surfaced as a number, so the feedback loop that justified granularity on Google does not close.
The correction. Three to five ad groups for a first build, split by buying situation rather than by phrasing. Let each accumulate enough spend that its cost per conversion means something. Granularity is a reward for having data, not a route to it.
Habit two: waiting for the search terms report
What the instinct says. Launch broad, let it run a week, read the search terms, add thirty negatives, reallocate. That loop is most of the job on Google and most of the value a manager adds in month one.
Why it fails here. There is no search terms report. You never see which conversation your ad appeared beside. There is also no negative keyword list, no exclusion field of any kind, so even if you could see waste you would have nowhere to put the exclusion.
Managers who wait for this report spend their first two weeks in a holding pattern, then discover the optimisation they planned is not available. The campaign has by then spent its learning budget on nothing.
The correction. Decide in week one that the only levers are ad group structure, hint wording, creative and landing page, and start moving them immediately. Treat the first two weeks as a creative test, not a query-mining exercise. If a low-fit theme is suspected, it gets its own ad group so it can be paused, which is the nearest available substitute for a negative.
Habit three: reaching for bid automation
What the instinct says. Manual CPC is a legacy setting. Put it on Target CPA or Maximise Conversions, feed the signal, get out of the way.
Why it fails here. ChatGPT Ads offers three objectives: Reach on CPM, Clicks on CPC, Conversions on oCPC. Bids are set at ad group level. This is closer to Google Ads in 2013 than 2026, and the automation being invoked is an optimisation layer, not a portfolio strategy you can tune.
The deeper problem is conversion volume. Any conversion-optimised bidding needs events to learn from. On a 25 dollar daily minimum, with CPC bid guidance circulating in the three to five dollar range as of 2026, reported by agency write-ups rather than published by OpenAI as a floor, a day buys something like five to eight clicks. A B2B account will not generate a learning dataset in a month.
The correction. Start on Clicks with a manual bid, run it until there is enough conversion volume to justify oCPC, and accept that this may take longer than a Google manager expects. The 60 dollar default max CPM figure circulating for Reach campaigns is likewise reported, not published.
Habit four: building the negative list on day one
What the instinct says. Import the master negative list before launch. Job seekers, students, free, DIY, jobs, salary, template. It costs nothing and saves waste.
Why it fails here. There is no field to put it in. This is the blank that surprises people most, because on Google the exclusion layer is so cheap it stops feeling like a control at all.
The correction. The negative list becomes ad copy. Every entry on it is an audience you want to repel, and the only place to repel them before you pay is the ad the user reads. A price floor removes the free-tool hunter. A headcount band removes the solo founder and the enterprise buyer at once. An integration requirement removes anyone on the wrong stack.
This lowers click-through rate by design, which is a real cost in a relevance-weighted auction and one nobody outside OpenAI can currently size. Hold the disqualifier back until there is enough spend to read, then judge it on cost per qualified lead rather than on CTR.
Habit five: judging week-one CTR against Google search benchmarks
What the instinct says. A 2 percent CTR on search is weak, 5 percent is healthy, under 1 percent means the ad is wrong. Those numbers are internalised to the point of being reflexes.
Why it fails here. They were formed on a surface where the user typed a query and scanned a results page containing four ads above ten links. ChatGPT Ads appear below a response that has already answered the question, labelled Sponsored and visually separated from the answer, per OpenAI's Ads in ChatGPT documentation. The user has what they came for before the ad is in view.
There is also no public benchmark to replace the Google one with. OpenAI publishes no cross-advertiser benchmarks, and nobody has released a category-level dataset. A manager who declares the channel dead at 0.4 percent in week one is comparing against a number from a different surface with no evidence the comparison holds.
The correction. Set the success criterion before launch, in cost per qualified lead, and hold it there. CTR on this channel is an input to the auction and a diagnostic for creative testing. It is not a verdict, because there is nothing credible to compare it against.
The one Google habit worth keeping
Account hygiene. UTM discipline, one destination URL convention, a conversion definition agreed with sales before launch, and a written record of what changed on which date. On a channel this young, with attribution windows only recently configurable and no search terms report to reconstruct history from, your own change log is the closest thing to an audit trail you will have.
What we cannot tell you
- What a normal CTR is on ChatGPT Ads. OpenAI publishes no cross-advertiser benchmarks and no category dataset exists. Absent by design.
- What relevance weighting is worth in cost terms. The mechanism is documented, the magnitude is not, because no quality score is surfaced.
- How long oCPC needs to learn. Undocumented. No conversion volume threshold is published.
- Whether three to five ad groups is actually optimal. This is our judgement from the reporting floor, not a measured result. No first-party InPromptAds data exists.
Quick answers
Should I use SKAG structure on ChatGPT Ads? No. The ad group is the smallest reporting unit, so single-theme granularity splits your data rather than sharpening it. Three to five ad groups split by buying situation is a workable first build.
When will the search terms report appear in my account? It will not. ChatGPT Ads has no search terms report at all. You never see the conversation your ad appeared beside, and no exclusion layer exists to act on it if you could.
Can I use automated bidding? Conversions campaigns run on oCPC, but any conversion optimisation needs events to learn from. At a 25 dollar daily minimum and reported three to five dollar CPC guidance, most B2B accounts will not reach useful volume quickly. Start on Clicks with a manual bid.
Where do my negative keywords go? Nowhere. There is no exclusion field. Move the intent of the list into ad copy as checkable disqualifying conditions, and split suspected low-fit themes into separately pausable ad groups.
Is a 0.5 percent CTR bad? Unknown. Google search benchmarks were formed on a different surface where the ad sits above the answer rather than below it, and no ChatGPT Ads benchmark has been published by anyone.
What transfers cleanly from Google Ads? Campaign and ad group hierarchy, location targeting, first-party audience lists as custom audiences, conversion tracking shape, and account hygiene. The auction is still a relevance-weighted second price family.
Sources
| Claim | Source | Tier |
|---|---|---|
| Campaign > ad group > ad structure; bid, destination URL and hints at ad group level | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Objectives are Reach (CPM), Clicks (CPC), Conversions (oCPC) | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Reporting at campaign, ad group and ad level only | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Ads are labelled Sponsored and visually separated from the response | OpenAI Help Center, Ads in ChatGPT, 2026 | Confirmed, primary |
| Minimum daily budget 25 USD | OpenAI Ads Manager documentation, 2026 | Confirmed, primary |
| Three to five dollar CPC and 60 dollar default max CPM guidance | Agency write-ups citing OpenAI guidance, 2026 | Reported |
| No search terms report, negatives, match types or visible quality score | OpenAI documentation, by absence | Absent |
| No cross-advertiser CTR benchmarks published | OpenAI reporting documentation, by absence | Absent |
| Three to five ad groups for a first build | InPromptAds | Inference, ours |
Related reading
- Mapping Your Google Ads Keyword Clusters to Context Hints
- Context Hints Are Not Keywords: The Translation Table
- Can You Exclude Conversations on ChatGPT Ads?
- How ChatGPT Ads Bidding Works
- ChatGPT Ads Measurement: What Actually Reports
Changelog
12 September 2026, v1.0. First publication. Names the five imported Google Ads habits that break on ChatGPT Ads and the correction for each.
Field kit
Tools
Site
OpenAI (primary)
Keshav studies how AI systems retrieve, verify, and cite brand information. At InPromptAds, he leads source research and turns platform documentation into practical guidance for advertisers.
